AI Visibility Audit

The prompt set is the instrument. It starts from questions buyers actually ask: sales call notes, support tickets, site search terms, Search Console queries phrased as questions and enquiry-form wording. Each prompt is classified by funnel stage, from problem awareness through comparison to the decision.

Five prompt types cover the space. Brand prompts ask about you by name and test accuracy. Category prompts ask who provides a service without naming anyone. Comparison prompts ask for alternatives or weigh two options. Problem prompts describe a need without naming a solution. Market prompts add a city, country or audience.

Language is a dimension of the set, not a translation step. Turkish and English prompts are written for each market's phrasing; Arabic and Persian can be added for brands serving those audiences, produced natively with reviewer approval. You sign off the final set, so the audit measures the questions you care about, not the easy ones.

AI answers change between runs, so one screenshot proves little. Every prompt is run several times, on different days within an agreed window, and each run is logged with its date, platform, mode and any visible model label. A finding counts as a pattern only when it recurs across runs; one-off answers stay in the log, flagged as variance.

Runs use clean sessions: logged out where the platform allows it, otherwise a fresh account with memory and personalisation off, no prior conversation, and location and interface language set to the target market. These conditions are written into the protocol, because a future re-run is only comparable if it follows the same rules. Where a tool such as Ahrefs Brand Radar covers the same prompts, it serves as a cross-check, not the reference.

Each answer is coded on the same fields. Presence: not mentioned, mentioned, or mentioned wrongly, with the wrong statement quoted. Position: named first, listed, or mentioned in passing. Citation: whether one of your pages is linked as a source. Competitors: every other brand named in the same answer. Sources: every cited domain and URL, which builds the source map.

Accuracy is checked against a fact sheet you approve at the start: services, locations, product names and founding facts. Tone records whether the framing is neutral, favourable or carries a caveat, such as an old complaint repeated from a forum. The distinction matters: an absent brand needs retrievable content, while a misdescribed brand needs its facts corrected at the sources the engine trusts.

Every gap is traced to the lowest layer that explains it, because content work is wasted if a crawler cannot reach the page. Access comes first: robots.txt rules per user agent, plus WAF and bot-protection settings that can block search crawlers even when robots.txt allows them. Rendering is next: if key text appears only after client-side JavaScript, some fetchers see an empty page. Then indexation: a page missing from Google or Bing cannot be retrieved by features built on those indexes.

The upper layers concern meaning. Entity clarity asks whether your name, category, locations and offer are stated consistently across your site, Organization markup, sameAs profiles and directories. Passage citability asks whether the page that should answer a prompt contains a self-contained passage that does so. Third-party coverage asks whether the domains the engines cite for your category mention you at all.